This analysis comes from Bloomberg Intelligence Senior Industry Analyst Mandeep Singh and Associate Analyst Nishant Chintala. It first appeared on Bloomberg Terminal.
Google Cloud could see at least $2 billion in revenue increase in 2025 from training and inference of generative AI workloads, which would help drive profitability faster than expected. The company appears to be well-positioned to leverage more of its Nvidia compute unit allocation for its enterprise customers compared to other hyperscalers like Meta and Amazon.com.
Empower your work with monthly insights into Enterprise Data & Technology
Sign up for the newsletter
Google Cloud could achieve a revenue increase of 400-500 bps
Although Alphabet has half the graphics processing units (GPUs) of Nvidia compared to hyperscalers like Meta and Microsoft, its internal use of GPU processing power for training and inference workloads is far lower, according to our calculation. Google Cloud already has annual revenue of $35 billion, with AI workloads expected to contribute at least 400-500 basis points to the segment’s growth in 2025, excluding possible revenue contributions from Duet AI Copilot and Gemini licenses and subscriptions.
Microsoft has seen a similar boost from AI inference workloads in its Azure segment. Given Microsoft’s close alliance with OpenAI, we believe that other base model companies do not train their large language models (LLMs) in the cloud.
Incremental headroom for AI workloads
Google’s operating profit growth is expected to accelerate, driven by the higher incremental margin of generative AI workloads. Google Cloud is already reporting an operating profit of $3.2 billion in 2024, compared to annual losses of $3 billion in 2021 and 2022. The company could have an advantage in gaining a share of genetic AI -Conquer workloads that require different types of computing power over others. those for conventional servers. We believe Google is well-positioned to close the gap with Amazon AWS and Microsoft Azure by leveraging its own TPU chip architecture, which could accelerate Google Cloud’s rise in profitability through 2024-25.
Nvidia GPU compute availability
Alphabet appears well-positioned to monetize more of its Nvidia GPU capacity for its customers in the cloud segment, compared to other hyperscalers that use the GPU compute capacity to train their own base models. Most of Google’s internal Gen AI workloads are optimized for performance on its own TPU accelerator chips. Google Cloud will likely offer a full suite of products for training large language models, which could cost up to $100 million and may become more expensive given the increasing complexity and parameters within these models.
Security could be a differentiator
Security could be another offering that sets Google Cloud apart from other hyperscalers as companies look to consolidate suppliers. Similar to how Microsoft bundles its security offerings with Azure cloud infrastructure, we expect Google Cloud adoption to increase due to the native security offered in the cloud. The company’s recent Cloud Next event unveiled a number of security features that we believe could become a differentiating factor as companies look to adopt generative AI.